Authors
Balyura O.V.1, Grebeniuk E.A.2, Eselevich R.V.1, Akbashev R.A.1, Surov D.A.1
1 S.M.Kirov Military Medical Academy, St. Petersburg
2 Saint-Petersburg State Marine Technical University, St. Petersburg
Abstract
Rationale. The choice of surgical treatment tactics for postoperative ventral hernias (PVH) is still based on a subjective visual assessment of computed tomography (CT) examinations by a surgeon, which leads to variability of solutions and lack of standardization. There are no approved software packages for automatic analysis of the anatomy of the anterior abdominal wall.
Objective. To develop a method for the automatic calculation of key surgical parameters (RDR, hernial sac volume) based on the semantic segmentation of the structures of the anterior abdominal wall using the convolutional neural network U-Net to objectify preoperative planning.
Methods. A retrospective single–center study was conducted based on data from 25 patients with PVH of categories W2-W3 (2024–2025). Manual segmentation of the right and left rectus abdominis muscles and hernial sac on CT sections was performed. The U-Net model was trained with a Focal Loss function and data augmentation. Based on the segmentation results, the algorithm automatically calculated the RDR (Rectus Diastasis Ratio) parameter and the volume of the hernial sac, forming a prognostic conclusion about the need for a component separation technique (CST).
Results. The model demonstrated stable learning (validation losses: 0,0015). The segmentation quality of the hernial sac was 74,9%. The automatic calculation of RDR allowed us to correctly classify treatment tactics for all patients: with RDR > 1,5, simple plastic surgery is recommended (40% of patients), with RDR <1,5 – CST (60% of patients). The analysis time is reduced to 2–3 minutes per patient versus 30-60 minutes for manual assessment.
Conclusion. The proposed method based on the U-Net convolutional neural network makes it possible to automatically identify key structures of the anterior abdominal wall on CT images. This creates the basis for the development of decision support systems capable of quantifying hernial defect parameters and objectifying preoperative planning.
Keywords: ventral hernia, deep learning, preoperative planning, semantic segmentation, artificial intelligence in surgery.
References
1. Bosanquet DC, Ansell J, Abdelrahman T, et al. Systematic Review and Meta-Regression of Factors Affecting Midline Incisional Hernia Rates: Analysis of 14,618 Patients. PLoS One. 2015; 10(9): e0138745. doi: 10.1371/journal.pone.0138745.
2. Pavelets KV, Vavilova OG, Lobanov MYu, et al. Features of preoperative preparation in patients with extensive and giant postoperative ventral hernias. Vestnik Sankt-Peterburgskoj medicinskoj akademii poslediplomnogo obrazovaniya. 2011; 3(2): 32-36. (In Russ.)
3. Timerbulatov ShV, Abdullin UM, Viktorov VV, et al. Intra-abdominal hypertension and abdominal compartment syndrome. Literature review. Kreativnaya xirurgiya i onkologiya. 2024; 14(2): 174-179. (In Russ). doi: 10.24060/2076-3093-2024-14-2-174-179.
4. Lukoyany`chev EE, Izmajlov SG, Evsyukov DA, et al. Possibilities of influencing the systemic level of inflammation in the acute phase after abdominal wall prosthetic surgery. Vestnik NMCzX im. N.I. Pirogova. 2022; 17(3): 53-58. (In Russ.) doi: 10.25881/ 2078255_2022_17_3_53.
5. Ermolov AS, Koroshvili VT, Blagovestnov DA. Postoperative ventral hernias – unresolved issues of surgical tactics. Zhurnal im. N.I. Pirogova. 2018; 10: 81-86. (In Russ.) doi: 10.17116/hirurgia201810181.
6. Kul`berg NS, Reshetnikov RV, Novik VP, et al. Variability in CT interpretation: one for all and all for one. Digital Diagnostics. 2021; 2(2): 105-118. (In Russ.) doi: 10.17816/DD60622.
7. Wang R, Mu Z, Wang J, et al. ASF-LKUNet: Adjacent-scale fusion U-Net with large kernel for multi-organ segmentation. Comput Biol Med. 2024; 181: 109050. doi: 10.1016/j.compbiomed.2024.109050.
8. Qin H, Long Y, Shengwei T, Cheng Z, YingYing Y, Lei Z. SEDyConv: Spatially enhanced multi-dimensional dynamic convolution for medical multi-organ segmentation in CTs. Knowledge-Based Systems. 2025; 323: 113707. doi: 10.1016/j.knosys.2025.113707.
9. Muysoms FE, Miserez M, Berrevoet F, et al. Classification of primary and incisional abdominal wall hernias. Hernia. 2009; 13(4): 407-14. doi: 10.1007/s10029-009-0518-x.
10. Nudnov NV, Korobov AV, Skachkov AA, et al. Evaluation of the performance of artificial intelligence in identifying degenerative diseases of the lumbosacral spine. Vestnik rentgenologii i radiologii. 2024; 105(1): 20-28. (In Russ.) doi: 10.20862/ 0042-4676-2024-105-1-20-28.
